Sam Altman Says the Singularity Is Here. Did Your Paid Ads Get the Memo?

On July 26, 2026, Sam Altman said on the Relentless podcast that “we’re in the singularity, it’s gradual and positive.” Business Insider also reported that several AI academics disputed whether that bar had been met, which is probably the right amount of philosophical caution to carry into a media plan: enough to avoid swallowing the headline whole, not enough to ignore the operating data sitting in front of us. For paid ads, the better question is narrower: if the singularity is “here,” where has it already changed the dashboard, the auction, or the thing being targeted? [1]

Three numbers are worth putting in the same tracker for the rest of 2026. On queries where AI Overviews appeared, Seer Interactive found paid CTR fell from 19.70% to 6.34% in a September 2025 multi-organization study covering 3,119 terms, 1.1 million paid impressions, and 42 organizations. ChatGPT ad CPMs were reported to have dropped from about $60 at the February 2026 pilot launch to about $25 by late April after self-serve access and Criteo involvement changed the buying route. And MediaPost’s Joe Mandese estimated intentional agent-targeted ad buying at $6 billion, while making clear that this is an editorial estimate rather than an audited market-size figure. [2][3][4]

That is a more useful translation of Altman’s singularity claim for paid ads than another argument about whether the word singularity is being used correctly. One signal hits search demand capture. One hits a new conversational ad surface. One raises the uncomfortable possibility that some paid media will be optimized toward non-human agents. Those are not the same problem, and treating them as one AI budget line is how reporting gets vague right when it needs to get cleaner.

AI Overviews Are Already a Paid Search Reporting Problem

The AI Overview number is the least glamorous and the hardest to hand-wave away. Seer’s study compared paid performance on queries where AI Overviews appeared and found CTR dropping from 19.70% to 6.34%. That is a relative decline of roughly 68%, and it lands exactly where paid search teams feel pain first: impression volume can still look alive while clicks quietly stop arriving. [2]

A lower CTR on AI Overview queries does not automatically prove those clicks disappeared forever. Some users may have been satisfied by the generated answer. Some may have clicked organic results. Some may have refined the search and converted later. The paid-search buyer’s problem is simpler: the ad unit is now sharing the results page with a machine-generated answer that can absorb attention before the buyer ever gets a chance to pay for the click.

The September 2025 timing matters. That dataset came before later AI search expansion, so it should not be treated as a final 2026 benchmark. It is better used as an early damage reading: when the generated answer appears, paid search behavior can change enough to make blended campaign averages misleading. If an account rolls AI Overview-affected queries into the same line as ordinary commercial queries, the buyer may end up explaining a CTR decline as creative fatigue, bid pressure, match-type drift, or landing-page mismatch when the surface itself changed.

The operational move is not complicated, but it is easy to postpone. AI Overview exposure needs its own query-level annotation wherever the team can approximate it. Brand, nonbrand, category, and competitor queries should not be judged only by their historical CTR baselines if the SERP now contains an answer box that did not exist in the same form. A weekly search report that separates affected and unaffected queries will tell a clearer story than another slide saying “AI search is changing behavior.”

Cheap ChatGPT CPMs Are Interesting, Not Yet Self-Explaining

The ChatGPT ad pricing story has the shape every early-market buyer recognizes: scarce pilot access, high minimums, then fast compression once more inventory and easier buying mechanics arrive. Digiday reported CPMs around $60 at the February 2026 pilot, when the minimum spend was about $250,000, and around $25 by late April after self-serve access and a Criteo partnership helped bring the minimum to $0. PPC Land also reported the $25 CPM level as OpenAI moved toward a broader auction model. [3][5]

That price drop is tempting. A buyer who remembers early TikTok inventory, retail media’s messy first-party promises, or the first wave of connected-TV self-serve buying will at least want a test cell. The risk is treating a low CPM as if it were already a low CPA. The reported ChatGPT pricing comes through advertiser and agency sources, not official OpenAI rate cards, and it varies by route and category. It is a market signal, not a guaranteed clearing price for every advertiser.

The larger problem is measurement. Search Engine Land reported that the ChatGPT ads pilot launched without pixel infrastructure, click IDs, or user-level data that would allow advertisers to connect ad clicks to downstream conversions. Optimum7 described similar early-test limitations around attribution and advertiser visibility. That means the channel can look cheap at the top of the funnel while still being hard to defend in a performance review. [6][7]

A mature paid-search or paid-social channel lets the buyer argue with familiar evidence: click IDs, conversion windows, holdout logic, pixel events, server-side events, modeled conversions, incrementality tests if the budget supports them. ChatGPT ads, at least in the early reports, do not yet give buyers that normal toolkit. So the correct test question is not “Are ChatGPT ads cheap?” It is “What can this test prove with the measurement we actually get?”

For now, that pushes ChatGPT ads closer to controlled experimentation than always-on performance buying. A team can still test message resonance, audience fit, assisted demand, branded search lift, or geography-level incrementality. What it should not do is let a $25 CPM wander into the same spreadsheet logic as a channel with deterministic conversion tracking. Cheap inventory becomes useful only when the buyer knows what kind of proof the channel is capable of producing.

Agent-Targeted Spend Belongs in the Plan, With a Warning Label

The $6 billion agent-targeted ad spend estimate is the strangest number in the set because it sounds too large to ignore and too soft to treat like audited market data. MediaPost’s Joe Mandese framed it as an estimate of intentional agent-targeted buying, combining virtual influencer spending with a fraud-removed subset. That caveat matters. This is not the same as an IAB-certified market-size report, and it should not be presented to finance as one. [4]

Still, the direction is not trivial. If advertisers are intentionally buying impressions meant for agents, assistants, synthetic personas, or AI-mediated decision systems, then the target entity has changed. The old binary was usually human audience versus invalid traffic. Agent-directed media sits in a more awkward middle: sometimes unwanted automation, sometimes a deliberate buyer or recommender proxy, sometimes a measurement object that needs validation before it can be optimized.

That is why DoubleVerify’s validation work matters as a market signal. MediaPost reported that DoubleVerify launched validation for AI-agent-directed traffic, which does not prove the size or effectiveness of the category but does show that non-human traffic is becoming something platforms, verification vendors, and buyers may need to classify rather than simply discard. [4]

The practical distinction is whether the agent is contamination, audience, or intermediary. If an automated agent is scraping, spoofing, or generating low-quality activity, it belongs in fraud review and exclusion logic. If it is part of a shopping, research, booking, or recommendation workflow, the media question changes: what signal should influence the agent, and how would the advertiser know that influence produced value? Most accounts are not set up to answer that cleanly yet.

Do Not Blend the Three Disruptions Into One AI Line Item

The mistake to avoid in Q3 and Q4 2026 is building one “AI ads” bucket and asking it to explain everything. AI Overviews change the search results page before the click. ChatGPT ads change the inventory source, price curve, and attribution standard. Agent-targeted buying changes the entity a campaign may be trying to influence. Those are different enough that a single budget label will hide more than it reveals.

• For AI Overview-affected search, track query cohorts, paid CTR, CPC, conversion rate, impression share, and any visible change in downstream branded demand.
• For ChatGPT inventory, track CPM by buying route, available attribution fields, test design, lift proxies, and whether the campaign can be evaluated without user-level conversion paths.
• For agent-directed media, track whether traffic is validated, whether the agent is intended or excluded, and whether the campaign objective is human persuasion, machine recommendation, or fraud control.

This is also where the singularity label has some limited use. It is not a planning framework by itself. It does not tell a buyer how much to spend, which platform will win, or whether an AI assistant will become a better customer than a human. It does, however, force a cleaner question: which part of the acquisition system changed?

If the answer is the dashboard metric, separate the query set. If the answer is the auction price, isolate the inventory and its proof standard. If the answer is the target entity, decide whether that entity is a valid audience, a recommender to influence, or traffic to exclude. That is a less dramatic conclusion than “paid ads are over,” but it is the one a media buyer can act on before the next budget meeting.

References

1. OpenAI CEO Sam Altman Says the Singularity Has Arrived, Business Insider, July 2026.
2. How Will AI Search Affect Paid Ads in 2026?, Search Influence.
3. 'Everything is coming down': ChatGPT ads are getting cheaper, Digiday, 2026.
4. The Media-Buying Singularity Is Almost Here, MediaPost.
5. ChatGPT ad CPMs drop to $25 as OpenAI races toward global auction, PPC Land, 2026.
6. ChatGPT ads pilot leaves advertisers without proof of ROI, Search Engine Land, 2026.
7. ChatGPT Ads in 2026: What the Early Tests Reveal, Optimum7, 2026.